CN113220983A - Deep learning-based item selection method and device - Google Patents
Deep learning-based item selection method and device Download PDFInfo
- Publication number
- CN113220983A CN113220983A CN202010081827.5A CN202010081827A CN113220983A CN 113220983 A CN113220983 A CN 113220983A CN 202010081827 A CN202010081827 A CN 202010081827A CN 113220983 A CN113220983 A CN 113220983A
- Authority
- CN
- China
- Prior art keywords
- article
- item
- list
- pair
- deep learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 71
- 238000010187 selection method Methods 0.000 title description 12
- 238000000034 method Methods 0.000 claims abstract description 57
- 238000012549 training Methods 0.000 claims abstract description 43
- 230000006399 behavior Effects 0.000 claims abstract description 42
- 238000005070 sampling Methods 0.000 claims description 52
- 239000010410 layer Substances 0.000 claims description 28
- 230000002159 abnormal effect Effects 0.000 claims description 25
- 238000012545 processing Methods 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 10
- 239000002356 single layer Substances 0.000 claims description 3
- 239000000047 product Substances 0.000 description 20
- 238000010586 diagram Methods 0.000 description 11
- 230000006870 function Effects 0.000 description 11
- 238000004422 calculation algorithm Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 8
- 238000004891 communication Methods 0.000 description 7
- 230000004913 activation Effects 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000002829 reductive effect Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 230000000670 limiting effect Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 230000002860 competitive effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000012467 final product Substances 0.000 description 1
- 238000007429 general method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000002147 killing effect Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Accounting & Taxation (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Finance (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method and a device for selecting products based on deep learning, and relates to the technical field of computers. One embodiment of the method comprises: collecting user behavior data, wherein the user behavior data comprise item lists corresponding to the search terms; constructing an article pair according to the articles in the article list to obtain an article pair list corresponding to each search term; and constructing a choice model based on a deep learning network according to the article pair list corresponding to each search word, and selecting articles according to the choice model. According to the embodiment, the object pairs are constructed, the selection model is constructed based on the constructed object pairs and the deep learning network, and selection is performed according to the selection model, so that model training data can be greatly expanded, and the coverage rate and the accuracy of selection results are improved.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a method and a device for selecting products based on deep learning.
Background
CPS (Cost Per Sales, i.e. pay Per sale) systems often involve a large number of items. CPS selection is to score mass CPS items and select the most competitive high-quality hot-market items. The current CPS product selection method mainly comprises the following steps: and selecting the product based on the mode of activity registration, data statistics and manual participation and selecting the product based on the mode of shallow learning. But these options have low coverage and accuracy.
Disclosure of Invention
In view of this, embodiments of the present invention provide an item selection method and apparatus based on deep learning, which can greatly improve coverage and accuracy of an item selection result.
To achieve the above object, according to an aspect of the embodiments of the present invention, there is provided a method for selecting a product based on deep learning, including:
collecting user behavior data, wherein the user behavior data comprise item lists corresponding to the search terms;
constructing an article pair according to the articles in the article list to obtain an article pair list corresponding to each search term;
and constructing a choice model based on a deep learning network according to the article pair list corresponding to each search word, and selecting articles according to the choice model.
Optionally, constructing an article pair according to the articles in the article list, and obtaining an article pair list corresponding to each search term, including:
for any search word, taking an article list corresponding to the search word as a target article list, or taking a list obtained by deleting at least part of articles in the article list corresponding to the search word as the target article list; and starting from the first item in the target item list, taking two adjacent unpaired items as an item pair, and obtaining an item pair list corresponding to any search term.
Optionally, deleting at least a part of the items in the item list, including: and deleting the articles of which the index values of the preset indexes in the article list are less than or equal to the index threshold value.
Optionally, the method further comprises determining a rating label for each item in the list of item pairs; constructing an article pair according to the articles in the article list, and obtaining an article pair list corresponding to each search term, wherein the method further comprises the following steps: and deleting the item pairs with the same preset grading label in the item pair list.
Optionally, the user behavior data further includes: click rate and exposure for each item in the item list;
determining a hierarchical label for each item in the list of item pairs, comprising: and determining the click rate of each item in the item pair list according to the click rate and the exposure, and determining the grading label of each item in the item pair list according to the click rate.
Optionally, determining the click rate of each item in the item pair list according to the click rate and the exposure according to the following formula:
where score represents the item's click rate, x identifies the item's click rate, y represents the item's exposure, and n represents a weighting factor.
Optionally, constructing an option model based on a deep learning network according to the item pair list corresponding to each search term, where the method includes:
taking all article pairs in the article pair list corresponding to each search word as a candidate sample set, and sampling the candidate sample set to obtain a training sample set; and constructing a choice model based on a deep learning network according to the training sample set.
Optionally, sampling the candidate sample set to obtain a training sample set, including:
determining a sampling probability for each item in the candidate sample set, determining a sampling probability for each item pair in the candidate sample set based on the sampling probability for each item, and sampling from the candidate sample set based on the sampling probability for each item pair.
Optionally, the sampling probability of each item in the candidate sample set is determined according to the following formula:
in the formula, P represents the sampling probability of the article, and y represents the exposure amount of the article.
Optionally, after sampling the candidate sample set to obtain a training sample set, the method further includes:
judging whether the index value of a preset index is abnormal or not for any article in the candidate sample set; if yes, the index value is subjected to standardization processing.
Optionally, the determining whether the index value of the preset index is abnormal includes:
determining the index score of the preset index according to the following formula, and if the index score is less than or equal to a score threshold value, judging that the index value of the preset index is not abnormal; otherwise, judging that the index value of the preset index is abnormal:
in the formula, a' represents an index score of the preset index, a represents an index value of the preset index, mean represents an average of the preset index, and σ represents a standard deviation of the preset index.
Optionally, before constructing the choice model based on the deep learning network, the method further includes:
for any article pair in the training sample set, judging whether the click rate of the previous article in the article pair is greater than that of the next article; if so, marking a first sample label for any article pair, otherwise marking a second sample label for any article pair.
Optionally, the deep learning network employs a single hidden layer.
According to a second aspect of the embodiments of the present invention, there is provided a selection device based on deep learning, including:
the data acquisition module is used for acquiring user behavior data, and the user behavior data comprises an article list corresponding to each search term;
the article combination module is used for constructing article pairs according to the articles in the article list to obtain an article pair list corresponding to each search term;
and the model building module is used for building an option model based on a deep learning network according to the article pair list corresponding to each search word and selecting articles according to the option model.
Optionally, the article combination module constructs an article pair according to the articles in the article list to obtain an article pair list corresponding to each search term, where the article pair list includes:
for any search word, taking an article list corresponding to the search word as a target article list, or taking a list obtained by deleting at least part of articles in the article list corresponding to the search word as the target article list; and starting from the first item in the target item list, taking two adjacent unpaired items as an item pair, and obtaining an item pair list corresponding to any search term.
Optionally, the article combination module is further configured to delete at least a part of the articles in the article list according to the following steps: and deleting the articles of which the index values of the preset indexes in the article list are less than or equal to the index threshold value.
Optionally, the article combining module is further configured to: determining the grading label of each article in the article pair list, and deleting the article pairs with the same preset grading label in the article pair list after constructing the article pairs according to the articles in the article list and obtaining the article pair list corresponding to each search word.
Optionally, the user behavior data further includes: click rate and exposure for each item in the item list;
the item combination module determines a hierarchical label for each item in the item pair list, including: and determining the click rate of each item in the item pair list according to the click rate and the exposure, and determining the grading label of each item in the item pair list according to the click rate.
Optionally, the item ranking module determines the click rate of each item in the item pair list according to the click rate and the exposure according to the following formula:
where score represents the item's click rate, x identifies the item's click rate, y represents the item's exposure, and n represents a weighting factor.
Optionally, the model building module builds an option model based on a deep learning network according to the item pair list corresponding to each search term, including:
taking all article pairs in the article pair list corresponding to each search word as a candidate sample set, and sampling the candidate sample set to obtain a training sample set; and constructing a choice model based on a deep learning network according to the training sample set.
Optionally, the sampling the candidate sample set by the model building module to obtain a training sample set includes:
determining a sampling probability for each item in the candidate sample set, determining a sampling probability for each item pair in the candidate sample set based on the sampling probability for each item, and sampling from the candidate sample set based on the sampling probability for each item pair.
Optionally, the model building module determines a sampling probability for each item in the candidate sample set according to the following formula:
in the formula, P represents the sampling probability of the article, and y represents the exposure amount of the article.
Optionally, after the model building module samples the candidate sample set to obtain a training sample set, the method further includes:
judging whether the index value of a preset index is abnormal or not for any article in the candidate sample set; if yes, the index value is subjected to standardization processing.
Optionally, the determining, by the model building module, whether an index value of a preset index is abnormal includes:
determining the index score of the preset index according to the following formula, and if the index score is less than or equal to a score threshold value, judging that the index value of the preset index is not abnormal; otherwise, judging that the index value of the preset index is abnormal:
in the formula, a' represents an index score of the preset index, a represents an index value of the preset index, mean represents an average of the preset index, and σ represents a standard deviation of the preset index.
Optionally, before the model building module builds the choice model based on the deep learning network, the model building module is further configured to:
for any article pair in the training sample set, judging whether the click rate of the previous article in the article pair is greater than that of the next article; if so, marking a first sample label for any article pair, otherwise marking a second sample label for any article pair.
Optionally, the deep learning network employs a single hidden layer.
According to a third aspect of the embodiments of the present invention, there is provided an electronic device for selecting items based on deep learning, including:
one or more processors;
a storage device for storing one or more programs,
when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided by the first aspect of the embodiments of the present invention.
According to a fourth aspect of embodiments of the present invention, there is provided a computer readable medium, on which a computer program is stored, which when executed by a processor, implements the method provided by the first aspect of embodiments of the present invention.
One embodiment of the above invention has the following advantages or benefits: by constructing the object pairs, constructing the choice model based on the constructed object pairs and the deep learning network, and selecting the choice according to the choice model, the model training data can be greatly expanded, and the coverage rate and the accuracy of the choice result are improved.
Further effects of the above-mentioned non-conventional alternatives will be described below in connection with the embodiments.
Drawings
The drawings are included to provide a better understanding of the invention and are not to be construed as unduly limiting the invention. Wherein:
FIG. 1 is a schematic main flow chart of a product selection method based on deep learning in an embodiment of the present invention;
FIG. 2 is a schematic flow chart diagram illustrating a deep learning based culling method in an alternative embodiment of the invention;
FIG. 3 is a schematic diagram of the major modules of a deep learning based election device, according to an embodiment of the present invention;
FIG. 4 is an exemplary system architecture diagram in which embodiments of the present invention may be employed;
fig. 5 is a schematic block diagram of a computer system suitable for use in implementing a terminal device or server of an embodiment of the invention.
Detailed Description
Exemplary embodiments of the present invention are described below with reference to the accompanying drawings, in which various details of embodiments of the invention are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the invention. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
According to one aspect of the embodiment of the invention, a deep learning-based selection method is provided.
Fig. 1 is a schematic main flow diagram of an item selection method based on deep learning in an embodiment of the present invention, and as shown in fig. 1, the item selection method based on deep learning includes: step S101, step S102, and step S103.
And S101, collecting user behavior data.
The user behavior data is data obtained by detecting user behavior. User behavior data may include data for searching, browsing, clicking, purchasing, placing orders, etc. behaviors. Optionally, the user behavior data includes item lists corresponding to the respective search terms. Of course, the user behavior data may also include data for evaluating various indicators of user behavior, such as the exposure of various items in the item list, the number of clicks made by the user on various items in the item list, and so forth.
The time span of the user behavior data can be selectively set according to actual conditions. For example, collecting user behavior data in a large time span, such as all behavior data of a user; and for example, collecting user behavior data over a small time span, such as only one month of user behavior data.
In practical application, a time span threshold may be set, so that the time span of the collected user behavior data is not less than the time span threshold, for example, 7 days, 30 days, and the like. For example, in some scenarios, the user behavior data is relatively sensitive to time, and for example, the difference between the user behavior data on weekends and weekdays may be large, the time span threshold may be set to 7 days, and when the data is collected, the user behavior data of at least the last 7 days is collected. By setting the time span threshold, the accuracy of the selection result can be improved.
And S102, constructing an article pair according to the articles in the article list to obtain an article pair list corresponding to each search term.
The method is used for constructing the object pairs based on the Pairwise algorithm and finally constructing the selection model through a deep learning method to select the objects. The method can greatly expand training data, clearly identify long-tail objects which cannot be covered by shallow learning, and meet the requirements of complex service scenes by using the nonlinear characteristics and high-precision characteristics of deep learning.
Optionally, constructing an article pair according to the articles in the article list, and obtaining an article pair list corresponding to each search term, including: for any search word, taking an article list corresponding to the search word as a target article list, or taking a list obtained by deleting at least part of articles in the article list corresponding to the search word as the target article list; and starting from the first item in the target item list, taking two adjacent unpaired items as an item pair, and obtaining an item pair list corresponding to any search term.
The two adjacent unpaired objects are taken as one object pair for deep learning, so that the significant difference and the non-significant difference between the objects can be fully learned, especially the long-tail objects with relatively low exposure, and the coverage rate and the accuracy of the selection result are improved.
Optionally, deleting at least a part of the items in the item list, including: and deleting the articles of which the index values of the preset indexes in the article list are less than or equal to the index threshold value. The preset index refers to a preset index, and can be selectively set according to actual conditions, such as setting as exposure, click rate, sales volume, and the like. The index threshold value may be selectively set according to actual conditions, for example, set to a fixed value, or determined according to the index value of a preset index of each article.
In an optional embodiment, the quartile of the index value of the preset index of all the articles in the article list corresponding to each search term is used as the index threshold. When the quartile is determined, the index values of the preset indexes of all the articles in the article list corresponding to each search term are sorted in a descending order and divided into four equal parts, and then the index value of the article at the first dividing point is taken as the quartile. In the field of e-commerce, when exposure is used as a preset index, the quartile is generally about 6.
And articles with the index value of the preset index in the article list being less than or equal to the index threshold value are deleted, so that the accuracy of the product selection result can be improved while the calculation amount is reduced.
Optionally, the method further comprises determining a rating label for each item in the list of item pairs; constructing an article pair according to the articles in the article list, and obtaining an article pair list corresponding to each search term, wherein the method further comprises the following steps: and deleting the item pairs with the same preset grading label in the item pair list.
The grading labels are used for distinguishing different article grades, and the names and the number of the grading labels can be selectively set according to actual conditions. For example, three classification tags of "equal," "medium," "difference," and four classification tags of "equal," "three," and "four" are set. The preset rating label refers to one rating label among a plurality of rating labels. Illustratively, when three ranking labels of "equal," "medium," "difference," etc. are set, after the item pair lists corresponding to the respective search terms are obtained, the ranking labels of two items in the same item pair may be "equal" and "equal," "medium" and "medium," "difference" and "difference," equal "and" difference, "medium" and "difference," etc. Assuming that the "difference equal" classification label is set as a preset classification label, after the item pair list corresponding to each search term is obtained, the item pairs of the "difference equal" and "difference equal" classification labels in the item pair list are deleted.
In this example, the step of determining the hierarchical label of each item in the item pair list may be performed before the step of constructing an item pair, or simultaneously with the step of determining the item pair list corresponding to the search term, or after obtaining the item pair list corresponding to each search term. And deleting the article pairs with the same preset grading label in the article pair list, so that the accuracy of the product selection result can be improved while the calculation amount is reduced.
The rating labels for each item may be determined empirically, or based on an index value for one or more of the indicators. Optionally, the user behavior data further includes: click rate and exposure for each item in the item list. Determining a hierarchical label for each item in the list of item pairs, comprising: and determining the click rate of each item in the item pair list according to the click rate and the exposure, and determining the grading label of each item in the item pair list according to the click rate.
The exposure amount refers to the number of times the item is seen by the user, and regardless of the page or scene, the exposure amount of the item is increased once as long as the item is seen by the user once. The click volume refers to the number of times an item is clicked by a user. Generally, the larger the click rate is, the more times the item is clicked with the same exposure amount is, and therefore the click rate can reflect the preference of the user to some extent. And determining the grading label of the article according to the click rate, so that the selection result is more in line with the preference of the user, and the accuracy of the selection result is improved.
In determining the click rate, the click rate of the item may be determined by dividing the click rate of the item by the exposure of the item. Optionally, determining the click rate of each item in the item pair list according to the click rate and the exposure according to the following formula one:
where score represents the item's click rate, x identifies the item's click rate, y represents the item's exposure, and n represents a weighting factor. The value of the weighting factor can be selectively set according to the actual situation, the weighting factor can be set to 1.1 under the normal situation, and those skilled in the art can also set to other values, such as 1.2, 0.8, etc., according to the actual situation. The weighting factors for different items may be the same or different. The summation of one of the equations refers to the summation of the corresponding values of the two items in the pair.
Compared with the mode of obtaining the click rate of the article by dividing the click rate of the article by the exposure of the article, the click rate determined by the formula is closer to the optimal sequencing solution of the application technology such as search. And performing subsequent step of constructing a product selection model for product selection based on the click rate determined by the formula, so that the accuracy of the product selection result can be improved.
And S103, constructing a choice model based on a deep learning network according to the article pair list corresponding to each search word, and selecting according to the choice model.
In this step, features may be extracted and feature engineering may be performed on the basis of relevant attributes such as category statistics (e.g., popularity, sales ranking, etc.), coupon attributes, etc. through static attributes of the item (e.g., price, commission, etc.), activity information (e.g., chunking, second killing, etc.), store data (e.g., sales volume, DSR (Detail Seller service Rating system), store level, etc.), statistical attributes of the item (e.g., sales volume, commission change, etc.), and so on. And then constructing a product selection method by a deep learning method. Deep learning forms a more abstract class or feature of high-level representation properties by combining low-level features to discover a distributed feature representation of the data. The specific method of Deep learning may be selected according to actual situations, such as DBN (Deep belief network), CNN (convolutional Neural Networks), RNN (recursive Neural Networks), and the like.
Continuously training parameters of each neural network through the nonlinear characteristic of deep learning, constructing a choice model, and finally predicting the probability of each item being clicked through the choice model obtained through the deep learning so as to select the items.
When the choice model is constructed based on the deep learning network, a batch training mode can be adopted for training, for example, all the article pairs under the same search word are substituted into the neural network for forward feedback, then the total difference (namely the difference between the clicked true probability and the predicted value of the article pair) is calculated, and error back propagation is carried out, so that the number of error back propagation times is greatly reduced.
The neural network for deep learning is divided into three layers: an input layer, a hidden layer, and an output layer. The loss function may be:
in the formula, SiRepresenting the real probability of clicking the item pair formed by the item i and the item j; siRepresenting the true probability of item i being clicked, SjRepresenting the true probability of the item j being clicked, and the calculation mode is shown in a formula five; si＞SjTime Sij1, otherwise Sij0; f () represents an activation function, the content in parentheses represents the variable of the activation function, a represents the input of each layer, namely the input of engineering information related to the object pair, and theta represents an activation threshold; k represents the number of input layers, and M represents the number of hidden units in the hidden layer; f. ofm() Denotes the activation function of the mth hidden unit, the content in parentheses denotes the variables of the activation function of the mth hidden unit, akRepresenting the input of the kth input layer, wmkInput layer to hidden layer weight of the m hidden unit representing the k input layer, bmInput layer to hidden layer bias, w, of the mth hidden unit representing the kth input layermDenotes the weight of the hidden layer to the output layer of the mth hidden unit of the kth input layer, and b denotes the hidden layer to output layer bias.
Optionally, constructing an option model based on a deep learning network according to the item pair list corresponding to each search term, where the method includes: taking all article pairs in the article pair list corresponding to each search word as a candidate sample set, and sampling the candidate sample set to obtain a training sample set; and constructing a choice model based on a deep learning network according to the training sample set. The candidate sample set is sampled to obtain the training sample set, so that the randomness of the samples and the probability of selecting high-index (such as exposure) articles can be increased, and the accuracy of the result of selecting the articles is improved.
Optionally, sampling the candidate sample set to obtain a training sample set, including: determining a sampling probability for each item in the candidate sample set, determining a sampling probability for each item pair in the candidate sample set based on the sampling probability for each item, and sampling from the candidate sample set based on the sampling probability for each item pair. In this example, the sampling probability of an item pair is the product of the sampling probabilities of two items in the item pair. The accuracy of the selection result can be further improved by determining the sampling probability of each article pair separately.
Optionally, the sampling probability of each item in the candidate sample set is determined according to the following formula two:
in the formula, P represents the sampling probability of the article, and y represents the exposure amount of the article. The accuracy of the product selection result can be obviously improved by adopting the formula in the example to determine the sampling probability of the product.
Optionally, after sampling the candidate sample set to obtain a training sample set, the method further includes: judging whether the index value of a preset index is abnormal or not for any article in the candidate sample set; if yes, the index value is subjected to standardization processing. The normalization process is to replace an abnormal value with a non-abnormal value, for example, with an average value, a median, etc. of a preset index. The abnormal index value is subjected to standardization processing, so that the influence of the abnormal value on the selection result can be avoided, and the accuracy of the selection result is improved.
The judgment basis of the abnormal value can be selectively set according to the actual situation. For example, if the click rate of an item suddenly increases without changing other conditions, the item is likely to have a brush line, and the click rate of the item may be determined as an abnormal value. Optionally, the determining whether the index value of the preset index is abnormal includes:
determining the index score of the preset index according to a third formula, and if the index score is less than or equal to a score threshold value, judging that the index value of the preset index is not abnormal; otherwise, judging that the index value of the preset index is abnormal:
in the formula, a' represents an index score of the preset index, a represents an index value of the preset index, mean represents an average of the preset index, and σ represents a standard deviation of the preset index. The score threshold value can be selectively set according to actual conditions, and is set to be 3.5, 5 and the like. The identification effect on the abnormal value can be greatly improved through the formula, and the accuracy of the final product selection result is improved.
Optionally, before constructing the choice model based on the deep learning network, the method further includes: for any article pair in the training sample set, judging whether the click rate of the previous article in the article pair is greater than that of the next article; if so, marking a first sample label for any article pair, otherwise marking a second sample label for any article pair. Illustratively, the first sample label is 1, representing a positive sample, and the second sample label is 0, representing a negative sample; alternatively, the first swatch is labeled 0, representing a negative swatch, and the second swatch is labeled 1, representing a positive swatch.
According to the invention, training samples are obtained based on the object pairs, and the sample data is distributed widely, so that the neural network with a single hidden layer can meet the training requirements of related methods. Thus, optionally, the deep learning network employs a single hidden layer. And a single-layer hidden layer is adopted, so that the calculation amount can be reduced on the basis of meeting the requirement on accuracy. In other alternative embodiments, the number of hidden units in the hidden layer is 15. The phenomenon of overfitting or low accuracy can occur when the number of the hidden units is higher than or lower than 15, and the overfitting phenomenon can be avoided on the basis of meeting the accuracy by adopting the 15 hidden units.
The selection method according to the embodiment of the present invention is exemplarily described below with reference to fig. 2. The scenario of the embodiment is a method for collecting behavior data of a user in CPS advertisement delivery related applications and finally constructing a selection from the behavior data of the user. The selection scene is complex, and the factors to be considered include various aspects, such as commission, price, label, style, color, sales volume, classification and other factors of the item, and also relate to various aspects of factors, such as sales promotion activity in which the item is recently registered, comprehensive rating of a merchant or a shop, brand awareness, current popular elements and other non-item factors. Of course, factors for practical commercial applications, such as e-commerce billing, banning categories, etc., need to be considered in addition to the algorithm, and these factors also interfere with the conventional selection algorithm, especially in part of temporary commercial policies.
The general method is that positive and negative samples are judged by collecting Click-Through-Rate (CTR) data of an article, and then a machine learning training set of the positive and negative samples is constructed according to the height of the CTR data. Because of the long tail phenomenon (also known as the long tail effect, i.e., most of the demand will be concentrated on the head), the exposure and click-through for most CPS items are very low. Statistically, since the computed CTR with a small flow rate is not the true CTR of the article, the articles belong to the articles whose positive and negative boundaries are fuzzy. The method easily causes that most of the objects with fuzzy positive and negative boundaries cannot participate in training, and is limited by the number of CTR data (because the objects with obvious positive and negative sample distinguishing are usually concentrated in a few object categories, and the positive and negative samples cannot be enriched by simply enlarging the sampled objects), and because the training samples are reduced, an accurate model or an excessively simple model cannot be really trained, the requirement of CPS complex personalized advertisement delivery cannot be really met. The embodiment provides a method based on a Pairwise algorithm (a method for constructing an article pair through the CTR size) to construct positive and negative samples, and finally an algorithm is constructed through a deep learning method to select articles. The method can greatly expand training data, clearly identify long-tail objects which cannot be covered by shallow learning, and meet the requirements of complex service scenes by using the nonlinear characteristics and high-precision characteristics of deep learning. Referring to fig. 2, the specific method is as follows:
step S201, collecting user behavior data within at least 7 days, wherein the user behavior data comprises: item lists corresponding to the search terms;
step S202, determining the click rate of each article in the article pair list according to the click number and the exposure number, and calculating the formula in the previous paragraph, which refers to the formula I;
step S203, determining the grading label of each item in the item pair list according to the click rate; marking the classified labels of 'difference and the like' on the articles with the CTR less than or equal to 0.023, marking the classified labels of 'medium' on the articles with the CTR more than 0.023 and the CTR less than 0.06, and marking the classified labels of 'medium' on the articles with the CTR more than or equal to 0.06;
step S204, constructing an article pair according to the articles in the article list to obtain a candidate sample set; the construction method is described in the related description, and is not repeated herein;
s205, sampling the candidate sample set to obtain a training sample set; the sampling method is described in the related description, and is not described in detail herein;
step S206, standardizing the abnormal index values of the preset indexes; the standardization processing method is referred to the related description above, and is not repeated herein;
step S207, constructing a choice model based on a deep learning network; the construction method is described in the foregoing, and is not described herein again.
The invention can eliminate the interference of the actual scene to the algorithm to simultaneously improve the coverage rate and the accuracy of the algorithm to the CPS selection.
The main problems existing at present in the traditional activity-based registration, data statistics and manual participation mode include:
1) compared with a large candidate set of articles, few articles can be selected by moving the articles and manually checking the articles;
2) the manual experience is not necessarily correct (the characteristics of the article can change individually with time, regions and scenes, which are all fields which cannot be reached by the manual experience);
3) most long-tail articles with hot-pin potential cannot be identified;
the method for selecting products based on shallow learning has the following problems:
1) personalized advertisement commercial activities, multi-dimensional article attributes, complex application scenes, personalized media and other factors of various practical comprehensive applications cannot be covered;
2) the distribution of the extracted features in the feature space is relatively sparse; items such as CPS tend to be in the billions, but items sold in a single day tend to be in the millions, so most items sold in the sales characteristic are 0, i.e., sparse distribution in data;
3) the performance of the classifier is greatly affected by static attributes of the training item (e.g., price changes, commission escalation, etc.);
4) the conditions of most real application scenes cannot be covered by positive and negative samples during off-line training, and most samples are distributed at fuzzy boundaries of the positive and negative samples;
5) the specific application scene individuation and the specific knowledge attribute exist in part of articles, and the characteristics cannot be identified by the conventional data mining algorithm;
6) the peculiarities of the CPS ecology, such as the bias of the CPS user's preferences for commissions, coupons, updates, etc. attributes from the actual item's revenue, can cause interference to the model.
The problems with both of the above methods result in poor coverage and accuracy in making the selections. According to the method, the object pairs are constructed, the selection model is constructed based on the constructed object pairs and the deep learning network, and selection is performed according to the selection model, so that model training data can be greatly expanded, and the coverage rate and the accuracy of selection results are improved.
The product selection method provided by the embodiment of the invention can be used for recalling default products related to searching and recommending, can also be combined with technologies such as searching, recommending and advertisement pushing, and can be applied to various fields such as CPS-related advertisement putting, searching and recommending, social e-commerce and media cooperation.
According to a second aspect of the embodiments of the present invention, there is provided an apparatus for implementing the above method.
Fig. 3 is a schematic diagram of main blocks of an option device based on deep learning according to an embodiment of the present invention, and as shown in fig. 3, an option device 300 based on deep learning includes:
the data acquisition module 301 is used for acquiring user behavior data, wherein the user behavior data comprises an article list corresponding to each search term;
the article combination module 302 is used for constructing article pairs according to the articles in the article list to obtain an article pair list corresponding to each search term;
and the model building module 303 is used for building an option model based on the deep learning network according to the item pair list corresponding to each search term, and selecting items according to the option model.
Optionally, the article combination module constructs an article pair according to the articles in the article list to obtain an article pair list corresponding to each search term, where the article pair list includes:
for any search word, taking an article list corresponding to the search word as a target article list, or taking a list obtained by deleting at least part of articles in the article list corresponding to the search word as the target article list; and starting from the first item in the target item list, taking two adjacent unpaired items as an item pair, and obtaining an item pair list corresponding to any search term.
Optionally, the article combination module is further configured to delete at least a part of the articles in the article list according to the following steps: and deleting the articles of which the index values of the preset indexes in the article list are less than or equal to the index threshold value.
Optionally, the article combining module is further configured to: determining the grading label of each article in the article pair list, and deleting the article pairs with the same preset grading label in the article pair list after constructing the article pairs according to the articles in the article list and obtaining the article pair list corresponding to each search word.
Optionally, the user behavior data further includes: click rate and exposure for each item in the item list;
the item combination module determines a hierarchical label for each item in the item pair list, including: and determining the click rate of each item in the item pair list according to the click rate and the exposure, and determining the grading label of each item in the item pair list according to the click rate.
Optionally, the item ranking module determines the click rate of each item in the item pair list according to the click rate and the exposure according to the following formula:
where score represents the item's click rate, x identifies the item's click rate, y represents the item's exposure, and n represents a weighting factor.
Optionally, the model building module builds an option model based on a deep learning network according to the item pair list corresponding to each search term, including:
taking all article pairs in the article pair list corresponding to each search word as a candidate sample set, and sampling the candidate sample set to obtain a training sample set; and constructing a choice model based on a deep learning network according to the training sample set.
Optionally, the sampling the candidate sample set by the model building module to obtain a training sample set includes:
determining a sampling probability for each item in the candidate sample set, determining a sampling probability for each item pair in the candidate sample set based on the sampling probability for each item, and sampling from the candidate sample set based on the sampling probability for each item pair.
Optionally, the model building module determines a sampling probability for each item in the candidate sample set according to the following formula:
in the formula, P represents the sampling probability of the article, and y represents the exposure amount of the article.
Optionally, after the model building module samples the candidate sample set to obtain a training sample set, the method further includes:
judging whether the index value of a preset index is abnormal or not for any article in the candidate sample set; if yes, the index value is subjected to standardization processing.
Optionally, the determining, by the model building module, whether an index value of a preset index is abnormal includes:
determining the index score of the preset index according to the following formula, and if the index score is less than or equal to a score threshold value, judging that the index value of the preset index is not abnormal; otherwise, judging that the index value of the preset index is abnormal:
in the formula, a' represents an index score of the preset index, a represents an index value of the preset index, mean represents an average of the preset index, and σ represents a standard deviation of the preset index.
Optionally, before the model building module builds the choice model based on the deep learning network, the model building module is further configured to:
for any article pair in the training sample set, judging whether the click rate of the previous article in the article pair is greater than that of the next article; if so, marking a first sample label for any article pair, otherwise marking a second sample label for any article pair.
Optionally, the deep learning network adopts a single-layer hidden layer, and the number of hidden units in the hidden layer is 15
According to a third aspect of the embodiments of the present invention, there is provided an electronic device for selecting items based on deep learning, including:
one or more processors;
a storage device for storing one or more programs,
when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided by the first aspect of the embodiments of the present invention.
According to a fourth aspect of embodiments of the present invention, there is provided a computer readable medium, on which a computer program is stored, which when executed by a processor, implements the method provided by the first aspect of embodiments of the present invention.
Fig. 4 illustrates an exemplary system architecture 400 of a deep learning based election method or a deep learning based election device to which embodiments of the present invention may be applied.
As shown in fig. 4, the system architecture 400 may include terminal devices 401, 402, 403, a network 404, and a server 405. The network 404 serves as a medium for providing communication links between the terminal devices 401, 402, 403 and the server 405. Network 404 may include various types of connections, such as wire, wireless communication links, or fiber optic cables, to name a few.
A user may use terminal devices 401, 402, 403 to interact with a server 405 over a network 404 to receive or send messages or the like. The terminal devices 401, 402, 403 may have installed thereon various communication client applications, such as shopping-like applications, web browser applications, search-like applications, instant messaging tools, mailbox clients, social platform software, etc. (by way of example only).
The terminal devices 401, 402, 403 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop portable computers, desktop computers, and the like.
The server 405 may be a server providing various services, such as a background management server (for example only) providing support for shopping websites browsed by users using the terminal devices 401, 402, 403. The backend management server may analyze and perform other processing on the received data such as the product information query request, and feed back a processing result (for example, target push information, product information — just an example) to the terminal device.
It should be noted that the deep learning based selection method provided by the embodiment of the present invention is generally executed by the server 405, and accordingly, the deep learning based selection device is generally disposed in the server 405.
It should be understood that the number of terminal devices, networks, and servers in fig. 4 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.
Referring now to FIG. 5, shown is a block diagram of a computer system 500 suitable for use with a terminal device implementing an embodiment of the present invention. The terminal device shown in fig. 5 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present invention.
As shown in fig. 5, the computer system 500 includes a Central Processing Unit (CPU)501 that can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM)502 or a program loaded from a storage section 508 into a Random Access Memory (RAM) 503. In the RAM 503, various programs and data necessary for the operation of the system 500 are also stored. The CPU 501, ROM 502, and RAM 503 are connected to each other via a bus 504. An input/output (I/O) interface 505 is also connected to bus 504.
The following components are connected to the I/O interface 505: an input portion 506 including a keyboard, a mouse, and the like; an output portion 507 including a display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, and a speaker; a storage portion 508 including a hard disk and the like; and a communication section 509 including a network interface card such as a LAN card, a modem, or the like. The communication section 509 performs communication processing via a network such as the internet. The driver 510 is also connected to the I/O interface 505 as necessary. A removable medium 511 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 510 as necessary, so that a computer program read out therefrom is mounted into the storage section 508 as necessary.
In particular, according to the embodiments of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method illustrated in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network through the communication section 509, and/or installed from the removable medium 511. The computer program performs the above-described functions defined in the system of the present invention when executed by the Central Processing Unit (CPU) 501.
It should be noted that the computer readable medium shown in the present invention can be a computer readable signal medium or a computer readable storage medium or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present invention, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present invention, however, a computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
The modules described in the embodiments of the present invention may be implemented by software or hardware. The described modules may also be provided in a processor, which may be described as: a processor comprising: the data acquisition module is used for acquiring user behavior data, wherein the user behavior data comprises: item lists corresponding to the search terms; the article combination module is used for constructing article pairs according to the articles in the article list to obtain an article pair list corresponding to each search term; and the model building module is used for building an option model based on a deep learning network according to the article pair list corresponding to each search word and selecting articles according to the option model. The names of the modules do not limit the modules themselves in some cases, for example, the data collection module can also be described as a module for building an option model based on a deep learning network.
As another aspect, the present invention also provides a computer-readable medium that may be contained in the apparatus described in the above embodiments; or may be separate and not incorporated into the device. The computer readable medium carries one or more programs which, when executed by a device, cause the device to comprise: collecting user behavior data, the user behavior data comprising: item lists corresponding to the search terms; constructing an article pair according to the articles in the article list to obtain an article pair list corresponding to each search term; and constructing a choice model based on a deep learning network according to the article pair list corresponding to each search word, and selecting articles according to the choice model.
According to the technical scheme of the embodiment of the invention, the model training data can be greatly expanded and the coverage rate and the accuracy of the selection result can be improved by constructing the object pairs, constructing the selection model based on the constructed object pairs and the deep learning network and selecting the selection according to the selection model.
The above-described embodiments should not be construed as limiting the scope of the invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions can occur, depending on design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (13)
1. A method for selecting products based on deep learning is characterized by comprising the following steps:
collecting user behavior data, wherein the user behavior data comprise item lists corresponding to the search terms;
constructing an article pair according to the articles in the article list to obtain an article pair list corresponding to each search term;
and constructing a choice model based on a deep learning network according to the article pair list corresponding to each search word, and selecting articles according to the choice model.
2. The method of claim 1, wherein constructing an item pair from the items in the item list, and obtaining an item pair list corresponding to each search term comprises:
for any search word, taking an article list corresponding to the search word as a target article list, or taking a list obtained by deleting at least part of articles in the article list corresponding to the search word as the target article list; and starting from the first item in the target item list, taking two adjacent unpaired items as an item pair, and obtaining an item pair list corresponding to any search term.
3. The method of claim 2, wherein deleting at least a portion of the items in the item list comprises: and deleting the articles of which the index values of the preset indexes in the article list are less than or equal to the index threshold value.
4. The method of claim 1, further comprising: determining a rating label for each item in the list of item pairs;
constructing an article pair according to the articles in the article list, and after obtaining an article pair list corresponding to each search term, further comprising: and deleting the item pairs with the same preset grading label in the item pair list.
5. The method of claim 4, wherein the user behavior data further comprises: click rate and exposure for each item in the item list;
determining a hierarchical label for each item in the list of item pairs, comprising: and determining the click rate of each item in the item pair list according to the click rate and the exposure, and determining the grading label of each item in the item pair list according to the click rate.
6. The method of claim 5, wherein constructing the option model based on the deep learning network according to the item pair list corresponding to each search term comprises:
taking all article pairs in the article pair list corresponding to each search word as a candidate sample set, and sampling the candidate sample set to obtain a training sample set; and constructing a choice model based on a deep learning network according to the training sample set.
7. The method of claim 6, wherein sampling the set of candidate samples to obtain a set of training samples comprises:
determining a sampling probability for each item in the candidate sample set, determining a sampling probability for each item pair in the candidate sample set based on the sampling probability for each item, and sampling from the candidate sample set based on the sampling probability for each item pair.
8. The method of claim 6, wherein after sampling the candidate sample set to obtain a training sample set, further comprising:
judging whether the index value of a preset index is abnormal or not for any article in the candidate sample set; if yes, the index value is subjected to standardization processing.
9. The method of claim 6, wherein prior to building the choice model based on the deep learning network, further comprising:
for any article pair in the training sample set, judging whether the click rate of the previous article in the article pair is greater than that of the next article; if so, marking a first sample label for any article pair, otherwise marking a second sample label for any article pair.
10. The method of claim 1, wherein the deep learning network employs a single layer of hidden layers.
11. An item selection device based on deep learning, comprising:
the data acquisition module is used for acquiring user behavior data, and the user behavior data comprises an article list corresponding to each search term;
the article combination module is used for constructing article pairs according to the articles in the article list to obtain an article pair list corresponding to each search term;
and the model building module is used for building an option model based on a deep learning network according to the article pair list corresponding to each search word and selecting articles according to the option model.
12. An electronic device for selecting items based on deep learning, comprising:
one or more processors;
a storage device for storing one or more programs,
the one or more programs, when executed by the one or more processors, implement the method of any of claims 1-10.
13. A computer-readable medium, on which a computer program is stored, which, when being executed by a processor, carries out the method according to any one of claims 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010081827.5A CN113220983A (en) | 2020-02-06 | 2020-02-06 | Deep learning-based item selection method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010081827.5A CN113220983A (en) | 2020-02-06 | 2020-02-06 | Deep learning-based item selection method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113220983A true CN113220983A (en) | 2021-08-06 |
Family
ID=77085555
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010081827.5A Pending CN113220983A (en) | 2020-02-06 | 2020-02-06 | Deep learning-based item selection method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113220983A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115169337A (en) * | 2022-08-24 | 2022-10-11 | 中教畅享(北京)科技有限公司 | Method for calculating keyword score in commodity title optimization |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103116639A (en) * | 2013-02-20 | 2013-05-22 | 新浪网技术(中国)有限公司 | Item recommendation method and system based on user-item bipartite model |
| CN107247798A (en) * | 2017-06-27 | 2017-10-13 | 北京京东尚科信息技术有限公司 | The method and apparatus for building search dictionary |
| CN107908662A (en) * | 2017-10-18 | 2018-04-13 | 北京京东尚科信息技术有限公司 | The implementation method and realization device of search system |
| WO2018192491A1 (en) * | 2017-04-20 | 2018-10-25 | 北京京东尚科信息技术有限公司 | Information pushing method and device |
| CN109426974A (en) * | 2017-08-25 | 2019-03-05 | 北京奇虎科技有限公司 | Competing product analysis method and device |
| CN110390052A (en) * | 2019-07-25 | 2019-10-29 | 腾讯科技（深圳）有限公司 | Search for recommended method, the training method of CTR prediction model, device and equipment |
-
2020
- 2020-02-06 CN CN202010081827.5A patent/CN113220983A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103116639A (en) * | 2013-02-20 | 2013-05-22 | 新浪网技术(中国)有限公司 | Item recommendation method and system based on user-item bipartite model |
| WO2018192491A1 (en) * | 2017-04-20 | 2018-10-25 | 北京京东尚科信息技术有限公司 | Information pushing method and device |
| CN107247798A (en) * | 2017-06-27 | 2017-10-13 | 北京京东尚科信息技术有限公司 | The method and apparatus for building search dictionary |
| CN109426974A (en) * | 2017-08-25 | 2019-03-05 | 北京奇虎科技有限公司 | Competing product analysis method and device |
| CN107908662A (en) * | 2017-10-18 | 2018-04-13 | 北京京东尚科信息技术有限公司 | The implementation method and realization device of search system |
| CN110390052A (en) * | 2019-07-25 | 2019-10-29 | 腾讯科技（深圳）有限公司 | Search for recommended method, the training method of CTR prediction model, device and equipment |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115169337A (en) * | 2022-08-24 | 2022-10-11 | 中教畅享(北京)科技有限公司 | Method for calculating keyword score in commodity title optimization |
| CN115169337B (en) * | 2022-08-24 | 2023-02-14 | 中教畅享(北京)科技有限公司 | Method for calculating keyword score in commodity title optimization |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108885624B (en) | Information recommendation system and method | |
| CN110473040B (en) | Product recommendation method and device and electronic equipment | |
| CN111444395A (en) | Method, system and equipment for obtaining relation expression between entities and advertisement recalling system | |
| CN105159910A (en) | Information recommendation method and device | |
| CN111429161B (en) | Feature extraction method, feature extraction device, storage medium and electronic equipment | |
| CN110766486A (en) | Method and device for determining item category | |
| CN111429214B (en) | Transaction data-based buyer and seller matching method and device | |
| CN113254795A (en) | Training method and device for recommendation model | |
| CN113220983A (en) | Deep learning-based item selection method and device | |
| CN113778979A (en) | Method and device for determining live broadcast click rate | |
| CN113822734A (en) | Method and apparatus for generating information | |
| CN116823410A (en) | Data processing method, object processing method, recommending method and computing device | |
| CN112348300A (en) | Method and device for pushing information | |
| CN113450172A (en) | Commodity recommendation method and device | |
| CN113327146A (en) | Information tracking method and device | |
| CN113450167A (en) | Commodity recommendation method and device | |
| CN112989276A (en) | Evaluation method and device of information push system | |
| CN113360816A (en) | Click rate prediction method and device | |
| CN113837843B (en) | Product recommendation method and device, medium and electronic equipment | |
| CN111831892A (en) | Information recommendation method, information recommendation device, server and storage medium | |
| CN112015970A (en) | Product recommendation method, related equipment and computer storage medium | |
| CN107357847B (en) | Data processing method and device | |
| CN111666481A (en) | Data mining method and device, computer readable medium and electronic equipment | |
| CN110689032A (en) | Data processing method and system, computer system and computer readable storage medium | |
| CN113516524B (en) | Method and device for pushing information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |